QUICS: Quantifying Uncertainty in

Integrated Catchment Studies

D 4.7 Tool to advise on using appropriate

river pollutant transport model.

Lead Partner: University of Sheffield

v1.0: 30th May 2017
Report Details

Title: Tool to advise on using appropriate river pollutant transport model

Deliverable Number (If applicable): D 4.7

Author(s): Vivian Camacho Suarez, James Shucksmith, Alma Schellart

Dissemination Level: Public

Document History

Version Date Status Submitted by Checked by Comment

0.1 04/04/2017 Draft First Vivian Camacho Alma Schellart Need to check and
Suarez update the references
0.5 15/05/2017 Second Alma Schellart James Shucksmith Some typos and
Draft changes in wording
suggested
0.7 28/05/2017 Third draft Alma Schellart Will Shepherd and Suggested tightening
Simon Tait of focus of the report
1.0 30/05/3017 Final Alma Schellart Will Shepherd

Acronyms and Abbreviations

ADE Advection Dispersion Equation

ADZ Aggregated Dead Zone
1D 1 Dimensional
2D 2 Dimensional

Acknowledgements

This project has received funding from the European Union’s Seventh
Framework Programme for research, technological development and
demonstration under grant agreement no 607000.
Executive Summary
This report describes the development and initial demonstration of a software tool
developed to enable evaluation of structural and parameter uncertainty in models to
describe pollutants in rivers. This tool can therefore assist in selecting appropriate river
pollutant transport models.
The tool currently comprises of a flexible modelling framework which enables comparison
of pollutant transport models of different complexity, as well as comparison of uncertainty
in different longitudinal dispersion coefficients. The tool can generate concentrations for
different combinations of inputs and models and calculates residuals by comparing model
outputs (e.g. peak concentrations) to reference values. In the current tool, these reference
values are the outputs from the more complex model (ADE 2D), the user can substitute
this by observed field data. The tool was developed in Python and coded in a flexible way
so that it can be adapted to suit the user’s needs. Other parameters, such as for example
hydraulic roughness, can be added to allow further investigation of its impact on
uncertainty. The tool can also cope with input of probability distributions instead of single
parameter values, in order to run Monte Carlo simulations. This deliverable report,
however, only describes the application of the tool to compare different model structures,
using rivers from the database described in Rutherford (1994).
Appendices to this report give an introduction to currently available river water quality
models and software; a summary of the sources of uncertainty in river quality modelling
and the definitions used throughout this report; and a description of the set-up and use of
the modelling framework code. The code is available in the dissemination area of the
QUICS project website www.quics.eu.

3
CONTENTS

Executive Summary .................................................................................................................... 3

1 Introduction ....................................................................................................................... 5
1.1 Partners Involved in Deliverable ............................................................................................. 5
1.2 Deliverable Objectives ........................................................................................................... 5
1.3 Background............................................................................................................................ 5
1.4 The tool: a modelling framework for evaluating structural and parameter uncertainty in
models to describe pollutant transport in rivers .................................................................................. 6
2 The modelling framework ................................................................................................... 7
3 Example demonstration of the modelling framework tool ................................................. 11
4 Conclusions ...................................................................................................................... 17
5 REFERENCES ..................................................................................................................... 18
APPENDIX A. River water quality models and available software .............................................. 21
APPENDIX B. Sources of uncertainty in river quality modelling .................................................. 26
APPENDIX C. Software Tool Code Description ........................................................................... 28

4
1 Introduction

1.1 Partners Involved in Deliverable

University of Sheffield.

1.2 Deliverable Objectives

To prepare a flexible software tool that can be used to elucidate the relationship between
different sources of uncertainty in surface water quality modelling for models of varying
structure and complexity.

1.3 Background
Maintaining good surface water quality standards for different uses (e.g. drinking,
recreation, ecological habitat) has become a challenging task due to stricter regulations,
anthropogenic influences and climate change impacts. Environmental modelling assists in
the assessment and improvement of surface water quality by simulating and predicting
water quality conditions that may not be otherwise obtained from field monitoring.
Modelling the transport of solutes in rivers is important for both, point and non-point source
pollution management. Point sources such as combined sewer overflows, industrial,
domestic or wastewater discharges may have adverse effects on river health. Similarly,
non-point source pollutants such as agricultural or urban runoff (for which quantification
may be more difficult due to their diffuse nature) can alter the natural ecosystems in rivers
leading to oxygen depletion and poor water quality conditions (Zheng et al., 2014).
Therefore, to understand and take pro-active actions towards good water quality
conditions, modelling offers a medium for water quality management. In order to model
river water quality processes and quantify the associated uncertainties it is necessary to
understand the natural river processes, modelling approaches, and the sources of
uncertainty in the modelling approach.
Quantifying and communicating the accuracy of model predictions is also a key
component for the proper management and decision making process (Refsgaard et al.,
2006; van Griensven and Meixner, 2006). Moreover, the European Water Framework
Directive suggests that water quality management should be addressed in an integrated
manner so a good ecological status of the water bodies is reached (EC, 2000). In this
integrated process, catchments are represented by models simplifying complex and non-
linear processes. The scientific community (e.g. Dotto et al., 2012; Beck, 1987), recognize
that modelling environmental systems can be highly uncertain, and there is a need to
assess the accuracy of surface water quality models in order to be able to improve their
predictions. Good quality high spatial and temporal resolution data is rare, which has made
it difficult to predict these processes in detail.
River water quality modelling uncertainties are often inherited from the physical,
mathematical and biochemical representation of pollutant transport processes and

5
transformation processes (structural uncertainty), quantification and selection of model
parameters such as mixing and hydraulic parameters (parameter uncertainty), data inputs,
and calibration (data input uncertainty).
This report will therefore describe the development of a modelling tool that can be used to
enable evaluation of structural and model parameter uncertainty as well as input data
uncertainty in models to describe pollutants in rivers. Appendices A, B, and C respectively
provide descriptions of the river water quality models and available software; common
sources of uncertainty; and the description of the modelling code developed for this
deliverable.

1.4 The tool: a modelling framework for evaluating structural and parameter
uncertainty in models to describe pollutant transport in rivers
This tool focuses on the model structures, model input data and parameters selected to
describe pollutant transport in rivers since these have been less studied and contribute
largely to the overall catchment uncertainty (Refsgaard et al., 2006; Zou et al., 2009).
Understanding how parameter and structural uncertainties behave over different spatial
and temporal scales provides a path towards uncertainty reduction in the modelling
process. To determine the dominant uncertainties in pollutant transport modelling in rivers,
the uncertainties due to the physical and biochemical model selection and parameter
uncertainty at the various temporal and spatial scales are currently being studied.
The initial phase of this work, as described in this deliverable, investigates the effect of the
model complexity on uncertainty over different spatial scales. By assessing how the
uncertainties respond to changes in river aspect ratios, the behaviour of structural
uncertainties can be studied.
A flexible modelling tool has been developed in the open source Python code
(https://www.python.org/). The tool can generate concentrations for different combinations
of input data and parameters and transport model structures and calculates residuals by
comparing each model outputs (e.g. peak concentrations) to reference values. In the
current framework, these reference values are the outputs from the more complex model
(e.g. ADE 2D), the user can substitute this by observed field data.

6
2 The modelling framework
Flexible software code was developed according to the framework shown in Figure 1, to
serve as a tool for uncertainty analysis considering a wide number of rivers with their
corresponding hydraulic and geomorphological characteristics. The tool is flexible to
implement several models and receive inputs in various forms, including probability
distributions. The tool calculates soluble pollutant concentrations for various models and
executes a structural and parameter uncertainty analysis. The framework contains data
input, modelling and uncertainty analysis modules (Figure 1). The tool receives and stores
the input parameters from a non-dimensional analysis for a specific river (e.g. narrow and
slow or wide and fast). A setup in the tool creates a rectangular river reach with the given
hydraulic and geomorphologic conditions. Consequently, the solute concentrations are
calculated and plotted within the space and time domains. Similarly, the peak
concentrations are calculated and plotted. The tool calculates the concentrations using the
various pollutant transport models (with varying complexities) including the advection only,
ADE and ADZ models. As part of future work, the tool will be extended to cover transient
storage and transformation models. This will enable a comparison of concentration
predictions from models with different complexities and parameter inputs.

Pollutant info River info Parametric Data

• Point source • Width, depth, shape • Dispersion coefficients
Input Data • Non point source • Flow, velocity, roughness • Travel time
• Conservative • Vegetation type • Delay time
• Non-conservative • Transient storage areas

Transport models
• Analytical ADE solutions
• ADZ
Modelling
• Transient Storage
• Decay
• Biochemical Reactions

Uncertainty Structural Uncertainty Parameter

Analysis Analysis Uncertainty Analysis

Figure 1. Modelling framework concept

7
Data input module
The data input module loads and stores the river, pollutant and mixing parameter variables
required by the various models. These variables include the river width, depth, length,
mean velocity, shear velocity, the pollutant mass, longitudinal and transverse dispersion
coefficients, mean travel time and residence time. The module is flexible so that these can
be implemented as deterministic values or as probability distributions. Both spatial and
temporal domains can be selected for analysis. The data input module can calculate either
the width or depth from the width to depth ratio and the mean velocity or shear velocity
from the mean to shear velocity ratio depending on the user specification. If the shear
velocity is not specified, the shear velocity is calculated using the relationship U* = √𝑔 ℎ 𝑆
where g is the gravitational constant, h is the depth and S is the channel slope. The tool
currently only handles steady uniform flow. The data input module can also calculate the
dispersion coefficients parameters from various methods as instructed by the user.

Modelling Module
To date, the module includes models describing the processes of (1) advection only in the
longitudinal direction, (2) advection-dispersion in the longitudinal direction (ADE 1D), (3)
advection and dispersion in both the longitudinal and transverse direction (ADE 2D), and
(4) aggregated dead zones. These models are briefly explained below with their analytical
solutions where applicable. In the future, the biochemical transformation models will also
be included in this tool.
Advection only assumes a pollutant is cross-sectionally well mixed and is based on the
river velocity as explained in Rutherford (1994) where the solute is transported only due to
the longitudinal mean velocity. This is also termed plug-flow, i.e., there is no change in the
concentration or spread of the pollutant.
Advection Dispersion Equation 1D (ADE 1D) – This equation is a depth and width
averaged form simplified from the three-dimensional advection dispersion equation,
assuming full instantaneous cross sectional mixing of a pollutant downstream of a release.
The Fischer et al. (1979) analytical solution to the ADE 1D equation is used, where the
pollutant concentration C(x,t) is:
Equation 1
𝑀 (𝑥 − 𝑉𝑥 𝑡)2
𝐶(𝑥, 𝑡) = exp ⌊− ⌋
𝐴 √4𝜋 𝐷𝑥 𝑡 4 𝐷𝑥 𝑡

Where M is the mass of the pollutant released at t=0 and x=0 and A is the cross sectional
area of the channel, Dx is the longitudinal dispersion coefficient, Vx is the average velocity
in the longitudinal direction, x is the spatial location and t is time. The advection dispersion
equation is widely used to predict solute concentrations within water quality modelling tools
(Kashefipour and Falconer, 2002).
Advection Dispersion Equation 2D (ADE 2D) – This equation is a depth averaged form of
the simplified form of the three-dimensional advection dispersion equation, assuming full
8
instantaneous mixing of a pollutant over the flow depth downstream of a release. This
model neglects transverse velocities (Vx>>Vy), but includes the effect of mixing in the
transverse direction. Concentrations are estimated using the analytical solution suggested
by Fischer et al (1979):
Equation 2
𝑀 (𝑥 − 𝑉𝑥 𝑡)2 𝑦2
𝐶(𝑥, 𝑦, 𝑡) = exp ⌊− − ⌋
4𝜋𝑑𝑡 √ 𝐷𝑥 𝐷𝑦 4 𝐷𝑥 𝑡 4 𝐷𝑦 𝑡

Where the transverse dispersion coefficient Dy and transverse dimension y are introduced
into the equation. The release of pollutant with mass M occurs at x=0, y=0 (at the middle
section of the stream width) and time t=0.
Aggregated Dead Zone (ADZ) – The ADZ model assumes that dispersion occurs mainly
due to dead zones. The model developed by (Beer and Young, 1983) estimates C using
the discretization
Equation 3

𝐶(𝑥2 , 𝑡) = 𝛼 𝐶(𝑥2 , 𝑡 − 1) + 𝛽 𝐶(𝑥, 𝑡 − 𝛿)

Where
∆𝑡
𝛼 = − exp ( ) 𝛽 = 1+ 𝛼
𝑇𝑅
𝛿 = 𝜏/∆𝑡 𝑇𝑅 = 𝑡̅ − 𝜏
TR is the residence time, 𝜏 (Tau) is the time delay, ∆𝑡 is the time step, and 𝑡̅ (Tbar) is the
average travel time.

Structural uncertainty analysis module

To determine the deficiencies in modelling due to the mathematical representations of the
models, it was initially assumed that the most complex model was the most accurate
representation of reality. Therefore, the ADE 2D model was used as baseline for
comparison of the other models. The modelling module obtains the estimated
concentrations from the advection only, ADE 1D and ADZ models and compares this
against the ADE 2D (and the framework has the flexibility to include other benchmarks).
The module then estimates the peak concentrations with respect to time. The peak
concentrations are normalized against the maximum concentrations and the models can
be compared. The uncertainty analysis module estimates the differences in each models’
estimations of peak concentrations in the river as a function of time. The model differences
are referred as residuals. Plots of the residuals against time or distance are created to
observe the spatial and temporal behaviour of the model results.

9
Parameter uncertainty analysis module
The tool is coded so that parameter uncertainty can be analysed using Monte Carlo
simulations, and the tool is flexible so that it can cope with input of both single values as
well as probability distributions.
As initial part of this study, a selection of methods for determining the longitudinal
dispersion coefficient (Dx) was obtained from: El Kadi Abderrezzak et al. (2015). These
include: Elder (1959), Fischer et al. (1979), McQuivey and Keefer (1974), Liu (1977),
Isawa and Aya (1991), Magazine et al. (1988), Koussis and Rodriguez Mirasol (1998), Seo
and Cheong (1998), Deng et al. (2001), Kashefipour and Falconer (2002) for B/H >50 (B/H
= width to depth ratio), and Kashefipour and Falconer (2002) for B/H <50. See Appendix B
for the full set of equations. Most of the methods for quantifying the longitudinal coefficient
are empirical relationships based on the aspect ratio (river width over depth) and
hydraulics of the river channel (mean and shear velocities). The initial part of this study
focusses on determining the impact of using the different longitudinal dispersion
coefficients on pollutant peak concentrations by comparing model predictions using the
different dispersion coefficient equations.
A comparison of the model predictions due to the inherent uncertainty in using
deterministic empirical dispersion coefficient equations by using Monte Carlo simulations
and probability distributions derived to describe uncertainty in Longitudinal Dispersion
Coefficients is currently being carried out (Camacho Suarez et al., In preparation).

10
3 Example demonstration of the modelling framework tool
Structural uncertainty analysis
Camacho et al. (2015) carried out an analysis to understand the impacts of differences in
structures of river mixing models on the estimated pollutant concentrations for rivers of
different characteristics. The Rutherford (1994) river database to classify rivers was used
to demonstrate the use of the tool for structural uncertainty analysis. Table 1 shows the
minimum, maximum and quartiles of the mean to shear velocity ratios and the aspect
ratios in the database. A wide river was defined as a river that had a large width with
respect to its depth, thus a large aspect ratio. Similarly, a fast river was defined as a river
with a large mean velocity and small shear stress resulting in a high mean to shear
velocity ratio. Peak concentrations were estimated for each model using the longitudinal
mixing coefficient from Elder (1959) and the transverse mixing coefficient from Fischer et
al. (1979). The parameters tau and mean travel time for the aggregated dead zone model
were obtained from the relationship given in Lees et al. (2000). For the uncertainty
analysis, the residuals for each model were calculated. The residuals were the result of the
differences in peak concentrations from the studied model and the complex model (ADE
2D). The assumption that the ADE 2D model was the most accurate model was made.
With the classification presented in Table 1, the modelling tool was used to calculate the
concentrations and the differences in models for a ‘wide’ versus a ‘narrow’ river, and a
‘fast’ versus ‘slow’ river (see Table 2).

Table 1. Distribution of mean to shear velocity and width to depth ratios obtained from Rutherford’s river
characteristics database
Vx/U* B/H
MIN 1.1 2.2
Q1 4.5 21.7
MEDIAN 6.1 37.9
Q3 9.0 60.0
MAX 21.0 174.0

Table 2 and Figure 2 show the differences between a narrow river (aspect ratio = 2.2) and
a wide river (aspect ratio = 173.8) which have similar mean to shear velocity ratios. The
modelling outputs (Figure 2) show the calculated concentrations of the models 50 m
downstream of the pollutant release at the centre of the cross section and the residuals
from the differences in peak concentrations and the more complex ADE 2D model. Figure
2 shows that in the narrow river, the advection only model over predicts the concentrations
while in the wide river, the advection only model under predicts when comparing it to the
ADE 2D. This over prediction of the advection only model in a narrow river is expected
because the advection only model assumes instantaneous mixing over the cross section,
and diluting the mass of the pollutant over a smaller cross section results in a higher
concentration as opposed to a wider cross section where the dilution occurs over a larger

11
area leading to a smaller concentration. This behaviour of over prediction in the narrow
river and under prediction in the wide river of the advection only is observed even at longer
time scales (observed in the residuals of the concentrations at the channel centreline as
shown in Figure 2). As opposed to the advection only model, the ADE 1D model has a
closer prediction to the ADE 2D model for the narrow river than the wide river where the
concentrations are under predicted. This indicates that for a narrow river, the ADE 1D and
the ADE 2D model predictions potentially converge faster than in wide rivers. This is
confirmed by looking at the residuals, where it can be seen that for the residuals decrease
with time and the models’ predictions converge albeit within 60 seconds in the narrow river
and 600 seconds in the wide river. This is expected because as the tracer travels
downstream in the ADE 2D model, transverse mixing occurs until the complete cross
section is mixed and hence conditions become similar to the ADE 1D model where
complete cross-sectional mixing is assumed at the pollution source. Once both models are
cross-sectionally well mixed, the models result in similar predictions and the structural
uncertainty decreases. The ADZ model under predicts the concentrations for both cases,
the wide and narrow river.

Table 2. Hydraulic and geomorphological properties of Rivers Yuma Mesa (‘narrow’) and Mississippi (‘wide’)
Yuma Mesa Mississippi
Depth (m) 3.45 3.1
Width (m) 7.6 530
Aspect ratio (-) 2.2 173.8
-1
Mean velocity (ms ) 0.68 0.08
-1
Shear velocity (ms ) 0.047 0.0056
Mean/shear velocity ratio (-) 14.47 14.2
Slope (-) 0.000065 0.000001
2 -1
Dx (m s ) 0.961 0.098
2 -1
Dy (m s ) 0.024 0.002
Pollutant mass (kg) 5.0 5.0
Tau (s) 125.9 1103.1
Tbar (s) 151.4 1311.8

12
Figure 2. Uncertainty tool outputs for Yuma Mesa and Mississippi Rivers: The Yuma Mesa River is a narrow
river where the advection-only model overestimates the tracer concentrations while the Mississippi River is a
wide river where the advection-only model under-predicts the concentrations. NB the plots show
concentrations at the channel centre line.

Similarly, the effect of the hydraulic conditions was analysed by comparing a slow to a fast
river. The slow river had a mean to shear velocity ratio of 1.24 while the fast river had a
mean to shear velocity ratio of 16.5, both rivers had similar aspect ratios (Table 3). Figure
3 shows that in both cases the advection only model over predicts when compared to the
ADE 2D in both slow and fast rivers. However, in the fast river, the differences between
the peak concentrations are smaller. This is expected since in faster and more turbulent
rivers, mixing in the transverse direction occurs more quickly.

13
Table 3. Hydraulic and geomorphological properties of Rivers Punehu (slow) and Coachella (fast)
Punehu Coachella
Depth (m) 0.28 1.56
Width (m) 5.0 24.0
Aspect ratio (-) 17.9 15.4
-1
Mean velocity (ms ) 0.26 0.71
-1
Shear velocity (ms ) 0.2098 0.0428
Mean/Shear velocity (-) 1.24 16.5
Slope (-) 0.0160 0.0001
2 -1
Dx (m s ) 0.395 0.395
2 -1
Dy (m s ) 0.008 0.010
Pollutant mass (kg) 5.0 5.0
Tau (s) 330.2 127.8
Tbar (s) 394.7 142.9

Figure 3. Modelling framework outputs for the Punehu (slower) River and Coachella (faster) River: The
Punehu River is a slower River than the Coachella River. In both rivers, the advection-only over predicts the
concentrations. NB the plots show concentrations at the channel centre line.

Initial work on Longitudinal Dispersion Coefficient Uncertainty Analysis

To understand the parameter uncertainty inherited from the longitudinal dispersion
coefficient (Dx) on the one-dimensional advection dispersion model (ADE 1D), ten
equations for calculating Dx were retrieved from El Kadi Abderrezzak et al. (2015). These
equations are listed in Table 4. One river was selected from the database (Table 5), and

14
the estimations for the ADE 1D model were computed using the different longitudinal
dispersion coefficients. Figure 4 shows the range of peak concentrations against time
obtained when using various dispersion coefficients. Elder’s longitudinal dispersion
coefficient (0.6 m2s-1) predicts higher peak concentrations than the other relationships of
dispersion coefficients. Elder’s equation estimates higher coefficients because it is
proportional to the multiplication of depth and shear velocity while some of the other
equations are inversely related to the shear velocity. In addition, Elder’s equation does not
include the transverse variations in velocity (El Kadi Abderrezzak et al., 2015). This leads
to a lower dispersion rate. On the cases of Seo and Cheong (1998), Deng et al. (2001),
and Kashefipour and Falconer (2002) equations, these have been obtained by regression
of larger data sets, resulting in more variables such as the shear to mean velocity ratio and
slope properties.

Table 4. Equations selected for initial work on longitudinal dispersion coefficient.

Model developed by Equation
Elder (1959) Dx = 5.93*h*Vs
Fischer (1975) Dx = 0.011 *(V/Vs)^2*(B/h)^2 *h*Vs
McQuivey and Keefer (1974) Dx = 0.058 *h*V/Se
Liu (1977) Dx = 0.18 * (V/Vs)^0.05 *(B/h)^2 *h*Vs
Isawa and Aya (1991) Dx = 2*h*Vs*(B/h)^1.5
Magazine et al. (1988) Dx = 75.86*(0.4*V/Vs)^-1.632 *Rh*V
Koussis and Rodriguez Dx = 0.6*h*Vs*(B/h)^2
Mirasol (1998)
Seo and Cheong (1998) Dx = 5.92 * (V/Vs) ^1.43 *(B/h) ^0.62 *h*Vs
Deng et al. (2001) Dx = (0.15/ (8 * Et)) *(V/Vs)^2 *(B/h) ^1.67 *h* Vs
Kashefipour and Falconer Dx = 10.612 * (V/Vs)* h* Vs
(2002) for B/h >50
Kashefipour and Falconer Dx = [7.428 +1.775 *(B/h) ^0.62 *(V/Vs)^ 0.572]* (V/Vs)* h* Vs
(2002) for B/h <50

Table 5. River and pollutant characteristics for longitudinal dispersion analysis

Value
Depth (m) 0.85
Width (m) 18.0
Aspect ratio (-) 21.2
-1
Mean velocity (ms ) 0.6
-1
Shear velocity (ms ) 0.10
Mean/Shear velocity (-) 6.0
Slope (-) 0.00120
2 -1
Dy (m s ) 0.0127
Pollutant mass (kg) 5.0

15
Figure 4. Peak concentrations for pollutant release using ADE 1D model and several longitudinal dispersion
2 -1
coefficients (Dx). Dispersion coefficients are in (m s )

The parameter uncertainty analysis will be expanded to include additional coefficients

derived from other studies, and to include the uncertainty inherent in using a deterministic
empirically derived Longitudinal Dispersion coefficient Dx (Camacho et al., in preparation).
Also the comparison with field collected dispersion coefficients will be carried out and
analysis of variables that affect the dispersion coefficient such as the Manning roughness
coefficient and the varying discharge flows.

16
4 Conclusions
A flexible tool was developed in Python, consisting of a framework in which different
mixing models and different model parameters can be included and the residuals
compared.
The framework has demonstrated the capability to run various models and estimate the
concentrations and the differences in model predictions. Therefore, the main tool for this
analysis has been developed and will continue to grow to accommodate additional
physical and biochemical models with their corresponding uncertainty analysis.

17
5 REFERENCES
Ani, E.-C., Wallis, S., Kraslawski, A., and Agachi, P. S.: Development, calibration and
evaluation of two mathematical models for pollutant transport in a small river,
Environmental Modelling & Software, 24, 1139-1152,
http://dx.doi.org/10.1016/j.envsoft.2009.03.008, 2009.
Beck, M. B.: Water quality modeling: A review of the analysis of uncertainty, Water
Resources Research, 23, 1393-1442, 1987.
Beer, T., and Young, P. C.: Longitudinal Dispersion in Natural Streams, Journal of
Environmental Engineering, 109, 1049-1067, doi:10.1061/(ASCE)0733-
9372(1983)109:5(1049), 1983.
Blumensaat, F., Seydel, J., Krebs, P., and Vanrolleghem, P. A.: Model structure sensitivity
of river water quality models for urban drainage impact assessment, Proceedings - 7th
International Congress on Environmental Modelling and Software: Bold Visions for
Environmental Modeling, iEMSs 2014, 2014, 609-618.
Camacho Suarez, V., Shucksmith, J.D., Schellart, A.N.A. (In preparation). Quantifying the
impact of uncertainty in the dispersion coefficient on water quality modelling in rivers.
Camacho, V., Norris, T., Shucksmith, J., Schellart, A., Analysis of Structural Uncertainty in
the Analytical Solution of ADE in River Impacted By CSOs, 10th International Conference
on Urban Drainage Modelling, Quebec, Canada, 203-207, 2015.
Choi, H. G., and Han, K. Y. P.: Development and applicability assessment of 1-D water
quality model in nakdong river, KSCE Journal of Civil Engineering, 18, 2234-2243,
10.1007/s12205-014-0457-7, 2014.
Cox, B. A.: A review of currently available in-stream water-quality models and their
applicability for simulating dissolved oxygen in lowland rivers, Science of The Total
Environment, 314–316, 335-377, http://dx.doi.org/10.1016/S0048-9697(03)00063-9, 2003.
Deng, Z. Q., Singh, V. P., and Bengtsson, L.: Longitudinal dispersion coefficient in straight
rivers, Journal of Hydraulic Engineering, 127, 919-927, 10.1061/(ASCE)0733-
9429(2001)127:11(919), 2001.
Dotto, C. B. S., Mannina, G., Kleidorfer, M., Vezzaro, L., Henrichs, M., McCarthy, D. T.,
Freni, G., Rauch, W., and Deletic, A.: Comparison of different uncertainty techniques in
urban stormwater quantity and quality modelling, Water Research, 46, 2545-2558,
http://dx.doi.org/10.1016/j.watres.2012.02.009, 2012.
EC: European Commision. Directive 2000/60/EC of the European Parliament and of the
Council of 23 October 2000 establishing a framework for Community actions in the field of
water policy, Official Journal of the European Communities, 327, 2000.
El Kadi Abderrezzak, K., Ata, R., and Zaoui, F.: One-dimensional numerical modelling of
solute transport in streams: The role of longitudinal dispersion coefficient, Journal of
Hydrology, 527, 978-989, 10.1016/j.jhydrol.2015.05.061, 2015.

18
Elder, J. W.: The dispersion of marked fluid in turbulent shear flow, Journal of Fluid
Mechanics, 5, 544-560, doi:10.1017/S0022112059000374, 1959.
Fischer, H. B., List, E.J., Koh, R.C.Y., Imberger, J, Brooks, N.H.: Mixing in Inland and
Coastal Waters, Academic Press, 1979.
Freni, G., and Mannina, G.: Bayesian approach for uncertainty quantification in water
quality modelling: The influence of prior distribution, Journal of Hydrology, 392, 31-39,
http://dx.doi.org/10.1016/j.jhydrol.2010.07.043, 2010.
Freni, G., Mannina, G., and Viviani, G.: Assessment of the integrated urban water quality
model complexity through identifiability analysis, Water Research, 45, 37-50,
http://dx.doi.org/10.1016/j.watres.2010.08.004, 2011.
González-Pinzõn, R., Haggerty, R., and Dentz, M.: Scaling and predicting solute transport
processes in streams, Water Resources Research, 49, 4071-4088, 2013.
Iwasa, Y., and Aya, S.: Predicting longitudinal dispersion coefficient in open-channel flows,
International Symposium on Environmental Hydraulics, Hong Kong, 505 – 510,1991.
Kannel, P.R., Kanel, S.R., Lee, S., Lee, Y.S., Gan, T.Y. A review of public domain water
quality models for simulating dissolved oxygen in rivers and streams. Environmental
Modelling and Assessment, 16, 2, 183-204, 2011.
Kashefipour, S. M., and Falconer, R. A.: Longitudinal dispersion coefficients in natural
channels, Water Research, 36, 1596-1608, http://dx.doi.org/10.1016/S0043-
1354(01)00351-7, 2002.
Koussis, A. D., and Rodríguez-Mirasol, J.: Hydraulic estimation of dispersion coefficient for
streams, Journal of Hydraulic Engineering, 124, 317-320, 1998.
Lees, M. J., Camacho, L. A., and Chapra, S.: On the relationship of transient storage and
aggregated dead zone models of longitudinal solute transport in streams, Water
Resources Research, 36, 213-224, 10.1029/1999WR900265, 2000.
Lindenschmidt, K.-E., Fleischbein, K., and Baborowski, M.: Structural uncertainty in a river
water quality modelling system, Ecological Modelling, 204, 289-300,
http://dx.doi.org/10.1016/j.ecolmodel.2007.01.004, 2007.
Liu, H.: Predicting dispersion coefficient of streams, ASCE J Environ Eng Div, 103, 59-69,
1977.
Magazine, M. K., Pathak, S. K., and Pande, P. K.: Effect of bed and side roughness on
dispersion in open channels, Journal of Hydraulic Engineering, 114, 766-782,
10.1061/(ASCE)0733-9429(1988)114:7(766), 1988.
Mannina, G., and Viviani, G.: A hydrodynamic water quality model for propagation of
pollutants in rivers, Water Science and Technology, 62, 288-299, 2010.
McQuivey, R. S., and Keefer, T. N.: Simple method for predicting dispersion in streams,
Journal of Environmental Engineering Division ASCE, 100, 997-1011, 1974.

19
Pelletier GJ, Chapra SC (2005) QUAL2Kw theory and documentation (version 5.1), a
modeling framework for simulating river and stream water quality. Washington:
Department of ecology
Refsgaard, J. C., van der Sluijs, J. P., Brown, J., and van der Keur, P.: A framework for
dealing with uncertainty due to model structure error, Advances in Water Resources, 29,
1586-1597, 10.1016/j.advwatres.2005.11.013, 2006.
Runkel, R. L., and Bencala, K. E.: Transport of reacting solutes in rivers and streams, in:
Environmental Hydrology, edited by: Singh, V. P., Kluwer Academic Netherlands, 137-164,
1995.
Rutherford, J. C.: River mixing, Wiley, 1994.
Seo, I. W., and Cheong, T. S.: Predicting longitudinal dispersion coefficient in natural
streams, Journal of Hydraulic Engineering, 124, 25-32, 1998.
Sharma, D. and Kansal, A. Assessment of river quality models: a review. Rev. Environ.
Sci. Biotechnol., 12, 285-311, 2013. DOI 10.1007/s11157-012-9285-8
Taylor, G.: The Dispersion of Matter in Turbulent Flow through a Pipe, Proceedings of the
Royal Society of London A: Mathematical, Physical and Engineering Sciences, 223, 446-
468, 10.1098/rspa.1954.0130, 1954.
Van Griensven, A. and Meixner, T.: Methods to quantify and identify the sources of
uncertainty for river basin water quality models, Water Sci. Technol., 53, 51–59, 2006
Wool TA, Ambrose RB, Martin JL, Comer EA (2001) Water quality analysis simulation
program, WASP6. Part A: model documentation, U.S. Environmental Protection Agency,
Center for exposure assessment modelling, Athens, GA
Wynne, B.: Uncertainty and environmental learning: Reconceiving science and policy in
the preventive paradigm, Global Environmental Change, 2, 111-127,
http://dx.doi.org/10.1016/0959-3780(92)90017-2, 1992.
Young, P. C., and Wallis, S. G.: Aggregated Dead Zone (ADZ) model for dispersion in
rivers, 1986, 421-433,
Zheng, Y., Han, F., Tian, Y., Wu, B., and Lin, Z.: Chapter 5 - Addressing the Uncertainty in
Modeling Watershed Nonpoint Source Pollution, in: Developments in Environmental
Modelling, edited by: Sven Erik Jørgensen, N.-B. C., and Fu-Liu, X., Elsevier, 113-159,
2014.
Zou, R., Lung, W.-S., and Wu, J.: Multiple-pattern parameter identification and uncertainty
analysis approach for water quality modeling, Ecological Modelling, 220, 621-629,
http://dx.doi.org/10.1016/j.ecolmodel.2008.11.021, 2009.

20
APPENDIX A. River water quality models and available software
River water quality modelling is mainly based on empirically calibrated mathematical
relationships derived from theoretical and empirical understanding of the key processes of
advection, dispersion and biochemical reactions.

Physical pollutant behaviour

In 1855, Adolf Eugen Fick made the analogy that salt diffuses in a water as heat diffuses in
a metal road. As a result, Fick’s law (Equation A1) stated that the net flux of tracer
concentration was proportional to the concentration gradient:
Equation A1
𝜕𝑐
𝐽𝑥 = −𝑒𝑚
𝜕𝑥
Where 𝐽𝑥 is the mass flux in (MT-1), −𝑒𝑚 is the molecular diffusion coefficient (L2T-1) and 𝑐
is the tracer concentration in (ML-3). Taylor (1954) showed that in pipe flow, there is a point
downstream (after the pollutant has been released) where the velocity shear and diffusion
reach equilibrium as discussed in the dispersion section (1.3.1). At this point, the variance
of the concentration profile increases linearly, the skewness reduces and the
concentrations tend towards a Gaussian distribution and hence can be described as a
Fickian process. Figure A1 shows the advective and equilibrium zones of a concentration
profile. The advective zone is the region closest to the source where the velocity
distribution plays a crucial role in mixing of the solute. The equilibrium zone is the zone
where the variance of the concentration profile becomes linear, the skewness reduces
(Rutherford, 1994) and the pollutant cloud can be modelled as a Fickian process. Figure
A1 can be expressed against either distance or time.

Figure A1. Variance and skewness vs. time of a change of concentration profile according to Fickian model
predictions (Rutherford, 1994)
Advection-Dispersion Equation (ADE)
The advection-dispersion equation is derived from the conservation of mass in a unit
volume where the accumulation of mass equals the mass input minus the mass output.
Assuming that the flow and cross section are constant, the advective flux is characterized
as the river velocity times the solute concentration.
The flux out of the volume is equal to the influx plus the change in flux in the control
volume. Then, Fick’s law (Equation A2) is used to describe the dispersive flux where the
mixing coefficient D is proportional to the concentration gradient. Both, advective and
dispersive fluxes are then placed in the conservation of mass equation leading to the 3D
ADE equation:
Equation A2
𝜕𝑐 𝜕𝑐 𝜕𝑐 𝜕𝑐 𝜕 𝜕𝑐 𝜕 𝜕𝑐 𝜕 𝜕𝑐
+ 𝑢𝑥 + 𝑢𝑦 + 𝑢𝑧 = (𝐷𝑥 ) + (𝐷𝑦 ) + (𝐷𝑧 ) + 𝑅(𝑐, 𝑃)
𝜕𝑡 𝜕𝑥 𝜕𝑦 𝜕𝑧 𝜕𝑥 𝜕𝑥 𝜕𝑦 𝜕𝑦 𝜕𝑧 𝜕𝑧
Where 𝑐 is the solute concentration in (ML-3), 𝑥 𝑦 𝑧 are the longitudinal, transverse and
vertical directions in (L), 𝑢𝑥 𝑢𝑦 𝑢𝑧 are the velocities (LT-1), 𝐷𝑥 𝐷𝑦 𝐷𝑧 are the mixing
coefficients in (L2T-1) and 𝑅 represents the biochemical transformations as a function of
concentration and other parameters (𝑃). The mixing coefficient represents mixing due to
both turbulence and diffusion, but they are usually combined into a single coefficient.

The three dimensional form of the ADE is the more complex method for estimating
concentrations. The ADE can be simplified assuming that the pollutant mixes
instantaneously across the depth, or over the width of the channel. In these cases, the
terms in the vertical or transverse dimensions can be eliminated. These simplifications are
reasonable in shallow rivers or rivers of narrow width. (Runkel and Bencala, 1995).

Aggregated Dead Zone (ADZ) model

The ADZ model is an alternative to the ADE. Developed by Young and Wallis (1986), the
ADZ considers the river an imperfectly mixed system (Figure A2) where advection takes
place first and then dispersion occurs in a mixing zone (Lees et al., 2000). Equation A3
shows the ADZ model where the change in concentration (𝑐) with respect to time (𝑡) is
proportional to the discharge 𝑄 in (L3 T-1) over the volume (𝑉) of water in the reach in (L3)
and the difference in upstream (𝑐𝑈 ) and downstream (𝑐𝐷 ) concentrations in (ML-3).

Figure A2. Representation of mixing zones in aggregated dead zone model (after Young and Wallis, 1986)

22
The ADZ model parameters include: (1) an average travel time which is the total time the
solute is advected and dispersed in the river, (2) a dispersive fraction that indicates the
ratio of mixed volume over the total volume, (3) a residence time that indicates the overall
time of travel associated with dispersion, and (4) a time delay due to the dispersive effect.
These parameters are presented under Equation A3:
Equation A3
𝜕𝑐(𝑡) 𝑄
= [𝑐𝑈 (𝑡) − 𝑐𝐷 (𝑡)]
𝜕𝑡 𝑉

Software for Water Quality Modelling

A variety of software for water quality modelling is available such as Simulation Catchment
(SIMCAT), QUAL2E, MIKE11, QUASAR, Flood Modeller Pro, SWAT, WASP, PEST, etc.
Most of the software available uses the one dimensional advection-dispersion equation
assuming full instantaneous mixing across the river cross-section. A synthesis of software
models reviewed is shown in Table A1.

23
Table A1: Summary of water quality software models
WATER DESCRIPTION INPUTS GOVERNING EQUATIONS DIME REVIEW DEVELOPED BY
QUALITY NSIO FROM
MODEL NS
Simulation Can model up to 600 reaches and Flow and quality data Does not include ADE, it assumes full 1D Kannel et Anglian Water
Catchment BOD, COD, DO, ammonia upstream, discharges and instantaneous mixing throughout each reach, al (2011),
(SIMCAT) concentrations abstractions (descriptions of solutes moving with same river velocity, uses B. A Cox
statistical distributions - means first order decay to calculate concentration that (2003)
and deviations) will enter next reach. Flow: mass balance
including tributary, effluent discharges,
abstractions, and upstream conditions. Solute:
for conservative pollutants - only advection, for
non-conservative - first order decay (BOD and
NH3), DO uses decay, temperature and
reaeration
Temporal Simulates current conditions and Physical parameters, flow and 1D Thames Water
overall model also changes to improve water quality data
for catchments quality. Can model storm events
(TOMCAT) by diverting effluent discharges

QUAL2EU It can model waste loads on in- River, global variables, and ADE 1D Kannel et US EPA
stream water quality and non-point forcing functions al (2011)
source waste loads

QUASAR To assess the environmental river map, boundary mass balance 1D developed as
impact of pollutants on river water conditions, observed data part of the
quality, real time control, dynamics Bedford Ouse
between the flow and water Study
quality, point and non-point
sources. pH, e Coli, algae, BOD,
DO, Conservative pollutants.

MIKE-11 flow and quality in rivers cross section, hydrodynamic St Venant equations (diffusive wave and B.A. Cox
advection dispersion and kinematic wave) (2003)
water quality parameters along
the river, flow and WQ time
series
WATER DESCRIPTION INPUTS GOVERNING EQUATIONS DIME REVIEW DEVELOPED BY
QUALITY NSIO FROM
MODEL NS
Flood modeller flow and quality boundary conditions, time St Venant equations, finite difference B.A. Cox
Pro (Former series of flow and water quality approximation to 1D- ADE, transformation (2003)
ISIS) at upstream and tributaries. equations
Also at downstream boundary,
cross section geometry,
abstractions, discharges
AQUATOX Nutrients, organic chemicals, Hydraulic, geometric, data, Mass balance of nutrients 1D Sharma, D.
suspended and bedded abiotic and biotic, physical, & Kansal,
sediments, DO fluctuations, biota, remineralisation and A. (2013)
toxicity from low oxygen and eco-toxicology
ammonia
One 16 variables including water Geometric data, initial 1D advection - dispersion with decay and sinks 1D Sharma, D.
dimensional temperature, N, P, DO, CBOD, conditions, model forcing data, & Kansal,
Riverine algae, Fe, MN, Coliform bacteria, hydraulic and control A. (2013)
Hydrodynamic macrophytes, varying point and parameters and calibration
and Water non-point source pollution, cycling data
Quality Model of nutrients, and fate, effect of
(EPD-RIV1) toxic materials
QUAL2Kw can model pathogens as a function flow and concentrations for In general first order decay, mass balance 1D Sharma, D. Pelletier &
of temperature, light settling headwater, discharges and & Kansal, Chapra (2005)
velocity, temp, pH, conductivity, withdrawals, reach segment A. (2013);
inorganic suspended solids, DO, lengths, elevations, hydraulic Kannel et
CBOD, N, Ammonia, P, biomass, geometry, weather data al 2011
algae, alkalinity,

WASP 7 Fate and transport of pollutants in Model segmentation, boundary ADE and kinetic transformation 1D, Sharma, D.
surface waters including DO, N, P, conditions, point and non-point 2D, & Kansal,
C, Temp, salinity, bacteria, silica, source loads, kinetic 3D A. (2013)
sediments, heavy metals, mercury, parameters, flow, initial (Wool
inorganic loads concentrations, numerical et al.
integration control options, 2001)
weather data
Water Quality water quality conditions in rivers geometry, meteorology, initial Conservation of heat and mass spatially and 1D Sharma, D.
for River - and reservoirs conditions, hydraulic and temporal, hydrologic routing, kinematic routing, & Kansal,
Reservoirs kinetic parameters steady flow, or full St Venat equations A. (2013)
Systems
Branched stream/river parameters, 1st order decays, 1D advective dispersion 1D Sharma, D.
Lagrangian global variable forcing equation (Langrangian reference frame) & Kansal,
Transport Model functions A. (2013)
(BLTM)

25
APPENDIX B. Sources of uncertainty in river quality modelling
Many different definitions of sources of uncertainty in models exist, however, this
report will use the terminology and focus on types of uncertainty described in Dotto
et al. (2012), as these are quantifiable sources of uncertainty in environmental
models. The sources considered are input data, model parameters, calibration data,
selection of objective functions and model structure uncertainties. Other sources of
uncertainty such as ‘ignorance’ as described by Wynne (1992) will not be the
included in this tool. Below, the definitions of structural and parameter uncertainties
that will be used throughout this toolkit are described.

Structural uncertainty
Model structure uncertainty is usually referred to as the uncertainty associated with
the deficiencies in matching the model to the real processes of interest (Refsgaard et
al., 2006). Frequently, in order to simplify complex processes, key components are
not considered or undergo scaling problems (Blumensaat et al., 2014). Model
structure uncertainty is also associated with the mathematical expressions chosen to
represent reality. Although widely accepted as a major source of uncertainty,
structural uncertainty has been often neglected (Refsgaard et al., 2006;Freni and
Mannina, 2010;Lindenschmidt et al., 2007).
Several sources of structural uncertainty are identified within the pollutant transport
and mixing models. For instance, the advection dispersion model does not represent
the skewness typically observed in tracer concentration profiles from field data.
Explanations including the effect of trapping areas (frequently called dead zones) or
the effect of shear velocities have been attributed to this skewness effect (e.g. data
collected in the advective zone). Other assumptions in the analytical solutions of the
ADE 1D such as an instantaneous fully mixed cross section of the pollutant mass
downstream of the source, or the steady uniform flow also lead to inaccuracies in the
predictions.
Another contribution to structural uncertainty is the reduction from the three-
dimensional space to one-dimensional space. Most solute transport studies involving
uncertainty analysis have been carried out in one dimension as observed in Mannina
and Viviani (2010) and Choi and Han (2014) and Ani et al. (2009). However, the
question of the level of uncertainty from dimensionality reduction still remains. It is
important to note that the inclusion of a second or third dimension involve the
inclusion of other parameters which will add to the parameter uncertainty (Ani et al.,
2009).

26
Parameter uncertainty
Parameter uncertainty is associated with the process of selection of the parameters
used in the model (Freni et al., 2011). The main parameters in pollutant transport
and mixing are the dispersion coefficients in the ADE model and time delay and
residence time in the ADZ model. These parameters are difficult and can be costly to
quantify since they often require field data collection for calibration.

ADE Dispersion Coefficients

The concept of longitudinal dispersion was first introduced by Taylor (1954) in a
circular pipe. Then, Elder (1959) derived a more theoretical dispersion equation for
an infinitely wide channel. Elder’s equation has been scrutinized for underestimating
the natural dispersion in rivers (Fischer et al, 1979). Since then, more empirical
equations for estimating the dispersion coefficients have been developed by
McQuivey and Keefer (1974), Liu (1977), Magazine et al. (1988), Seo and Cheong
(1998) and Kashefipour and Falconer (2002) amongst others. Through tracer
experiments, dispersion coefficients have also been estimated. But although these
experiments are more site specific and may give better parameter estimations, they
are also associated with field collection errors and are specific to the time of year
they were carried out, i.e. during high/low flow and summer/winter vegetation.

ADZ time delay and residence time

The time delay and residence time are obtained from the pollutant cloud travelling
times. The time delay is the advective time it takes for the cloud to move only due to
advection of the bulk flow while the residence time is a lumped parameter that
describes the travel time associated with dispersion (González-Pinzõn et al., 2013).
These parameters can also be obtained from experimental studies. Relationships
between the ADZ parameters and hydraulic conditions have been studied for
instance by Lees et al. (2000).

27
APPENDIX C. Software Tool Code Description
Overview
The code is developed in Python 2.7.1 (https://www.python.org) using the Spyder 2.3
platform (https://pypi.python.org/pypi/spyder/2.3.0), it is available from the
dissemination pages of the QUICS project website, www.quics.eu. On a higher level,
the code is separated into Module classes (Figure C1). Objects must be created.
Therefore, there are two ways to create the main module (CMain) which is the core
of the modelling framework. The first option is using the ‘1_framework’ file, and the
second option is using the ‘3_interface’ file. These are the wrappers to the CMain
class. The 1_framework allows the user to understand the code implementation. On
the other hand, the 3_interface wrapper gives the user a menu that can be used to
run the code directly. It also allows the user to load into the memory a list of analyses
that have already been saved. This way, the user can re-plot calculated data without
having to recalculate that data first. Both 1_framework and 3_interface are simply
different entry points to the same underlying code and its functionality.
The main module is the blueprint of the framework. It defines the:
- CInputModule: responsible for all input actions.
- CCalculateConcentrationsModule: responsible for all calculations of
concentrations.
- CPostPostprocessingModule: responsible for post-processing actions, such
as those made on the already-defined input parameters, or concentration that
may have been calculated.
- COutpuTModule: responsible for all output actions.
Many of these classes have methods to calculate values rather than storing them.
This is designed to avoid duplication. For example, the cross-sectional area is
defined as the result of width multiplied by depth. The area of a river is not stored as
this duplicates river data and if we should change either the depth or width, then we
either have to remember to update the area too, or risk getting our variables out of
sync with each other.
However, the modules do not contain the data. The data is contained in the
‘analyses’ object which is a list of analyses. Each analysis holds all the information to
define that analysis uniquely. Each permutation of input parameters constitutes a
new analysis. So if the same river is analysed twice, only with different spatial
parameters such as xStep = 0.1 and then xStep = 0.2, there will two separate
analysis objects.
This way, the framework simply iterates through each river in the input file (which is
defined in the CInputModule), and tells the main module to create a new analysis
object for each river and append it to its list of analyses. It continues to run through
all the steps in turn to calculate and process more data (e.g. concentrations) by

28
various means and plot the data in various forms. At the end, there is an option to
save the analyses to file.
In addition there is 2_TestFramework, which also provides an additional entry point
to run the same code, but not with the same purpose. 1_framework and 3_interface
provided different ways that the user can interact with the code. 2_TestFrameworkis
simply to test individual parts of the code to check that it has been implemented in
such a way that we satisfy unit tests.

Modules
CMainModule: Creates the ‘input’, ‘output’, ‘post-processing’, and ‘calculate
concentrations’ modules with their corresponding functions.
CInputModule: Requires the user to enter the following:
 Database of rivers containing the river names, depths (m), widths (m), mean
velocities (ms-1), and slopes.
 Longitudinal distance along the river (river_length) of interest (m).
 Mass of the pollutant released in the river (kg).
 Longitudinal and transverse dispersion coefficients (m2s-1). At the moment,
the code uses (Elder, 1959) coefficients, but this can be changed by entering
the dispersion coefficients in lines 73 and 74 of the input module.
CCalculateConcentrationsModule: Calculates the concentrations using the following
functions:
 Advection Only
 ADE_Longitudinal_Dispersion (ADE 1D)
 ADE_Transverse_Mixing (ADE 2D)
 ADZ
Then the concentrations are saved into the analysis object
COutputModule:
 Plots the concentrations vs time or distance.
 Calculates peak concentrations.
 Calculates normalised concentrations and plots them against the Peclet
number.
 Calculates peak normalised concentrations and plots them against the Peclet
number.

29
  It can calculate if the concentration of a water quality parameter has been
exceeded – This requires changing the default exceedance concentration
value of 2.0 by the value of interest.
CPostPostprocessingModule:
 Calculates the residuals. At the moment, the ADE 2D is considered the most
accurate model, therefore, the residuals are calculated using the ADE 2D as
the reference value, but this can be changed in the
PeakConcentrationsOverTimeResiduals function in the CModel class.
Analysis Object
The analysis object contains the river properties (CRiver), time and space domains
(CDomain), the calculated concentrations for the Advection only, ADE 1D, ADE 2D
and ADZ models (Cmodels) and the pollutant data (Cpollutant)
CRiver:
 The river class stores the river name, depth, width, flow, mean velocity, and
slope.
 It calculates the shear velocity, mean to shear velocity ratio, width to depth
ratio and cross-sectional area.
 It prints both, the river input and calculated properties to the console.
CDomain:
 Linearly discretises the x, y space and time domains.
 Prints the x, y and time steps to the console.
CPollutant:
 Stores the initial pollutant mass or the pollutant concentration.
 Prints the pollutant properties to the console.
CMixingParameters:
 Sets the default dispersion longitudinal and transverse dispersion coefficients.
 Prints the dispersion coefficients to the console.
Cmodels:
 Stores the calculated concentrations from the advection only, ADE 1D, ADE
2D and ADZ functions.
 Stores peak and normalised concentrations.
CSpatialParameters
 Stores the x, y time steps.

30
 1_FRAMEWORK/
3_INTERFACE

MAIN MODULE
CMain Analyses

INPUT MODULE CALCULATE OUTPUT MODULE POST PROCESSING MODULE Analysis

CInputModule CONCENTRATIONS COutputModule CPostProcessingModule
CCalculateConcentrationsModule CAnalysis

Input river Print Analyses to Domain River Models Pollutant Mixing Parameters
ADVECTION ONLY Calculate residuals
database console CDomain CRiver CModels CPollutant CMixingParameters

Define x index to
print Longitudinal
Input pollutant info ADE 1D X discretisation Width Concentrations Mass Dispersion
concentrations vs.
width Coefficient

Input dispersion Plot Concentrations Transverse

ADE 2D Peak
coefficients vs. time Y discretisation Depth Dispersion
Concentrations
Coefficient

Plot peak
Input x and y steps ADZ concentrations vs. Time discretisation Area Residuals
time

Plot normalized
concentrations vs. Width to depth
Peclet number ratio

Plot peak
normalized
concentrations vs. Shear velocity
Peclet number

Mean velocity

Figure C1. Pollutant Transport Modelling Framework Structure. The blue section is the blueprint for carrying out the calculations, while the r section
stores the data inputs, calculations and outputs

31
How to run the code:
1) Define inputs in the CInputModule:
a. Input river database in the form of River Name, Depth, Width, Shear Velocity
and Slope.
b. Define longitudinal distance or river length (river_length)
c. Define pollutant mass.
d. Define x and y steps.
e. Define longitudinal (x) and transverse (y) dispersion coefficients.
2) Run 1_framework or 3_interface.
3) Save the results if needed for later analysis.
4) Run 3_interface to load previous saved analysis.